sulagnasaharasha commited on
Commit
d218e37
·
verified ·
1 Parent(s): cec1ae4

Add detailed dataset card

Browse files
Files changed (1) hide show
  1. README.md +266 -83
README.md CHANGED
@@ -1,86 +1,269 @@
1
  ---
2
- dataset_info:
3
- - config_name: closeup
4
- features:
5
- - name: polygon_id
6
- dtype: int64
7
- - name: split
8
- dtype: string
9
- - name: species_label
10
- dtype: string
11
- - name: canopyrs_object_id
12
- dtype: int64
13
- - name: final_plant_name
14
- dtype: string
15
- - name: gbif_accepted_scientific_name
16
- dtype: string
17
- - name: area
18
- dtype: float64
19
- - name: habit
20
- dtype: string
21
- - name: closeup
22
- dtype: image
23
- splits:
24
- - name: train
25
- num_bytes: 700685099
26
- num_examples: 1336
27
- - name: val
28
- num_bytes: 140719427
29
- num_examples: 267
30
- - name: test
31
- num_bytes: 154392724
32
- num_examples: 294
33
- download_size: 995712416
34
- dataset_size: 995797250
35
- - config_name: temporal
36
- features:
37
- - name: polygon_id
38
- dtype: int64
39
- - name: date
40
- dtype: string
41
- - name: split
42
- dtype: string
43
- - name: species_label
44
- dtype: string
45
- - name: canopyrs_object_id
46
- dtype: int64
47
- - name: final_plant_name
48
- dtype: string
49
- - name: gbif_accepted_scientific_name
50
- dtype: string
51
- - name: area
52
- dtype: float64
53
- - name: habit
54
- dtype: string
55
- - name: crownview
56
- dtype: image
57
- splits:
58
- - name: train
59
- num_bytes: 6837373385
60
- num_examples: 21376
61
- - name: val
62
- num_bytes: 1320390188
63
- num_examples: 4272
64
- - name: test
65
- num_bytes: 1394965700
66
- num_examples: 4704
67
- download_size: 9550323640
68
- dataset_size: 9552729273
69
  configs:
70
- - config_name: closeup
71
- data_files:
72
- - split: train
73
- path: closeup/train-*
74
- - split: val
75
- path: closeup/val-*
76
- - split: test
77
- path: closeup/test-*
78
- - config_name: temporal
79
- data_files:
80
- - split: train
81
- path: temporal/train-*
82
- - split: val
83
- path: temporal/val-*
84
- - split: test
85
- path: temporal/test-*
86
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ license: cc-by-4.0
3
+ task_categories:
4
+ - image-classification
5
+ - image-to-image
6
+ language:
7
+ - en
8
+ tags:
9
+ - biodiversity
10
+ - remote-sensing
11
+ - tropical-forest
12
+ - tree-species
13
+ - aerial-imagery
14
+ - drone
15
+ - multi-temporal
16
+ - crown-view
17
+ - closeup
18
+ - BCI
19
+ - Panama
20
+ pretty_name: BCI Temporal Crown Dataset
21
+ size_categories:
22
+ - 10K<n<100K
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
23
  configs:
24
+ - config_name: temporal
25
+ data_files:
26
+ - split: train
27
+ path: temporal/train/*.parquet
28
+ - split: val
29
+ path: temporal/val/*.parquet
30
+ - split: test
31
+ path: temporal/test/*.parquet
32
+ - config_name: closeup
33
+ data_files:
34
+ - split: train
35
+ path: closeup/train/*.parquet
36
+ - split: val
37
+ path: closeup/val/*.parquet
38
+ - split: test
39
+ path: closeup/test/*.parquet
40
  ---
41
+
42
+ # BCI Temporal Crown Dataset
43
+
44
+ A multi-temporal, multi-modal dataset of **tropical tree crowns** from Barro Colorado Island (BCI), Panama. Each tree is observed across **16 acquisition dates** spanning June 2024 – September 2025, paired with a ground-level close-up photograph.
45
+
46
+ ---
47
+
48
+ ## Dataset Summary
49
+
50
+ | | |
51
+ |---|---|
52
+ | **Site** | Barro Colorado Island (BCI), Smithsonian Tropical Research Institute, Panama |
53
+ | **Tree crowns** | 1,897 labeled polygons across 84 species |
54
+ | **Raster dates** | 16 (monthly, June 2024 – September 2025) |
55
+ | **Total temporal rows** | ~30,000 (1,897 crowns × 16 dates) |
56
+ | **Crown area** | 7 – 1,212 m² (median ~160 m²) |
57
+ | **Image resolution** | 512 × 512 px, RGBA (alpha = crown mask) |
58
+ | **Growth form** | All freestanding trees |
59
+
60
+ ---
61
+
62
+ ## Configurations
63
+
64
+ This dataset has two configurations that can be **joined on `polygon_id`** at load time.
65
+
66
+ ### `temporal` — Crown-view tiles (one row per crown × date)
67
+
68
+ Each row is a masked aerial crown tile extracted from a monthly RGB orthomosaic raster.
69
+
70
+ | Column | Type | Description |
71
+ |---|---|---|
72
+ | `polygon_id` | int | Unique crown identifier (join key) |
73
+ | `date` | string | Acquisition date `YYYYMMDD` |
74
+ | `split` | string | `train` / `val` / `test` |
75
+ | `species_label` | string | Species name used as the classification label |
76
+ | `gbif_accepted_scientific_name` | string | GBIF-accepted full scientific name |
77
+ | `final_plant_name` | string | Field-verified plant name |
78
+ | `canopyrs_object_id` | int | Original CanopyRS object ID |
79
+ | `habit` | string | Growth form (all `Freestanding`) |
80
+ | `area` | float | Crown polygon area in m² |
81
+ | `crownview` | Image | 512×512 RGBA masked aerial tile |
82
+
83
+ **Size:** ~30,350 rows (train ~21,380 · val ~4,272 · test ~4,704 — 16 dates each)
84
+
85
+ ### `closeup` — Ground-level close-up photos (one row per crown)
86
+
87
+ Each row is a zoom photograph taken from a drone at lower altitude, centered on the crown.
88
+
89
+ | Column | Type | Description |
90
+ |---|---|---|
91
+ | `polygon_id` | int | Unique crown identifier (join key) |
92
+ | `split` | string | `train` / `val` / `test` |
93
+ | `species_label` | string | Species name used as the classification label |
94
+ | `gbif_accepted_scientific_name` | string | GBIF-accepted full scientific name |
95
+ | `final_plant_name` | string | Field-verified plant name |
96
+ | `canopyrs_object_id` | int | Original CanopyRS object ID |
97
+ | `habit` | string | Growth form |
98
+ | `area` | float | Crown polygon area in m² |
99
+ | `closeup` | Image | 512×512 RGBA center-cropped/padded close-up photo |
100
+
101
+ **Size:** 1,897 rows (train 1,336 · val 267 · test 294)
102
+
103
+ ---
104
+
105
+ ## Data Splits
106
+
107
+ Splits are **stratified by species** using a 70 / 15 / 15 allocation. Species with ≤ 6 crowns use fixed small-sample allocations to ensure representation across splits where possible.
108
+
109
+ | Split | Crowns | Species |
110
+ |---|---|---|
111
+ | train | 1,336 | 84 |
112
+ | val | 267 | 65 |
113
+ | test | 294 | 81 |
114
+ | **Total** | **1,897** | **84** |
115
+
116
+ ---
117
+
118
+ ## Species Distribution (Top 10)
119
+
120
+ | Species | Total crowns |
121
+ |---|---|
122
+ | *Anacardium excelsum* | 257 |
123
+ | *Dipteryx oleifera* | 190 |
124
+ | *Luehea seemannii* | 109 |
125
+ | *Prioria copaifera* | 95 |
126
+ | *Jacaranda copaia* | 90 |
127
+ | *Hieronyma alchorneoides* | 83 |
128
+ | *Virola surinamensis* | 63 |
129
+ | *Hura crepitans* | 57 |
130
+ | *Tachigali panamensis* | 45 |
131
+ | *Quararibea stenophylla* | 44 |
132
+
133
+ The dataset is **long-tailed**: 84 species total, many with < 10 crowns.
134
+
135
+ ---
136
+
137
+ ## Temporal Coverage
138
+
139
+ 16 monthly acquisition dates:
140
+
141
+ ```
142
+ 2024-06-11 2024-07-16 2024-08-13 2024-09-18
143
+ 2024-10-14 2024-11-12 2024-12-16 2025-01-24
144
+ 2025-02-17 2025-03-17 2025-04-14 2025-05-12
145
+ 2025-06-16 2025-07-15 2025-08-18 2025-09-15
146
+ ```
147
+
148
+ Dates span the **dry season** (January–April) and **wet season** (May–December) of the Panamanian tropics, capturing phenological variation.
149
+
150
+ ---
151
+
152
+ ## Image Details
153
+
154
+ ### Crown-view tiles (`crownview`)
155
+ - **Source**: RGB COG rasters acquired over BCI (~10 cm/px GSD)
156
+ - **Processing**: Each labeled crown polygon is tilerized using [geodataset](https://github.com/canopyrs/geodataset). Pixels outside the crown polygon are **zeroed out** (alpha = 0 in RGBA). Images are center-cropped or zero-padded to 512 × 512.
157
+ - **Format**: PNG-encoded RGBA, stored as HuggingFace `Image` feature
158
+
159
+ ### Close-up photos (`closeup`)
160
+ - **Source**: Drone zoom photos collected via the CanopyRS platform (`zoom_url` field)
161
+ - **Processing**: Downloaded from CanopyRS, center-cropped / zero-padded to 512 × 512 RGBA
162
+ - **Format**: PNG-encoded RGBA, stored as HuggingFace `Image` feature
163
+ - **Temporal note**: One close-up per crown (date-invariant) — join to `temporal` on `polygon_id`
164
+
165
+ ---
166
+
167
+ ## Usage
168
+
169
+ ### Load a single config
170
+
171
+ ```python
172
+ from datasets import load_dataset
173
+
174
+ # Multi-temporal crown views
175
+ temporal = load_dataset("sulagnasaharasha/bci-temporal", "temporal")
176
+ print(temporal["train"][0])
177
+ # {'polygon_id': 12345, 'date': '20250915', 'species_label': 'Anacardium excelsum',
178
+ # 'crownview': <PIL.Image ...>, ...}
179
+
180
+ # Close-up photos
181
+ closeup = load_dataset("sulagnasaharasha/bci-temporal", "closeup")
182
+ print(closeup["train"][0])
183
+ # {'polygon_id': 12345, 'species_label': 'Anacardium excelsum',
184
+ # 'closeup': <PIL.Image ...>, ...}
185
+ ```
186
+
187
+ ### Join temporal + closeup for multi-modal training
188
+
189
+ ```python
190
+ from datasets import load_dataset
191
+ import pandas as pd
192
+
193
+ temporal = load_dataset("sulagnasaharasha/bci-temporal", "temporal")
194
+ closeup = load_dataset("sulagnasaharasha/bci-temporal", "closeup")
195
+
196
+ # Convert to pandas and join
197
+ t = temporal["train"].to_pandas()
198
+ c = closeup["train"].to_pandas()[["polygon_id", "closeup"]]
199
+ paired = t.merge(c, on="polygon_id")
200
+ # Each row now has both crownview (date-specific) and closeup (date-invariant)
201
+ ```
202
+
203
+ ### PyTorch Dataset example
204
+
205
+ ```python
206
+ import torch
207
+ from torch.utils.data import Dataset
208
+ from datasets import load_dataset
209
+ from torchvision import transforms
210
+
211
+ class BCITemporalDataset(Dataset):
212
+ def __init__(self, split: str = "train", transform=None):
213
+ temporal = load_dataset("sulagnasaharasha/bci-temporal", "temporal", split=split)
214
+ closeup = load_dataset("sulagnasaharasha/bci-temporal", "closeup", split=split)
215
+
216
+ t_df = temporal.to_pandas()
217
+ c_df = closeup.to_pandas()[["polygon_id", "closeup"]]
218
+ self.df = t_df.merge(c_df, on="polygon_id").reset_index(drop=True)
219
+
220
+ self.species = sorted(self.df["species_label"].unique())
221
+ self.label_map = {s: i for i, s in enumerate(self.species)}
222
+ self.transform = transform or transforms.ToTensor()
223
+
224
+ def __len__(self) -> int:
225
+ return len(self.df)
226
+
227
+ def __getitem__(self, idx: int) -> dict:
228
+ row = self.df.iloc[idx]
229
+ crown = self.transform(row["crownview"].convert("RGB")) # [3, H, W]
230
+ closeup = self.transform(row["closeup"].convert("RGB")) # [3, H, W]
231
+ label = self.label_map[row["species_label"]]
232
+ return {"crownview": crown, "closeup": closeup,
233
+ "label": torch.tensor(label), "date": row["date"],
234
+ "polygon_id": row["polygon_id"]}
235
+ ```
236
+
237
+ ---
238
+
239
+ ## Source Data
240
+
241
+ - **Site**: [Barro Colorado Island (BCI)](https://stri.si.edu/facility/barro-colorado-island), Smithsonian Tropical Research Institute, Republic of Panama
242
+ - **Crown polygons**: Produced by [CanopyRS](https://canopyrs.org) using automated segmentation + expert annotation
243
+ - **Aerial rasters**: Monthly RGB orthomosaics acquired over BCI (COG format), hosted by the CanopyRS platform
244
+ - **Taxonomy**: Species names resolved against [GBIF Backbone Taxonomy](https://www.gbif.org/dataset/d7dddbf4-2cf0-4f39-9b2a-bb099caae36c) and [WCVP](https://wcvp.science.kew.org/)
245
+
246
+ ---
247
+
248
+ ## License
249
+
250
+ [Creative Commons Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/)
251
+
252
+ ---
253
+
254
+ ## Citation
255
+
256
+ If you use this dataset, please cite:
257
+
258
+ ```bibtex
259
+ @dataset{saharasha2025bcitemporal,
260
+ author = {Saharasha, Sulagna},
261
+ title = {{BCI Temporal Crown Dataset}: Multi-temporal aerial crown tiles
262
+ paired with ground-level close-up photos for tropical tree
263
+ species recognition},
264
+ year = {2025},
265
+ publisher = {HuggingFace},
266
+ url = {https://huggingface.co/datasets/sulagnasaharasha/bci-temporal},
267
+ note = {Barro Colorado Island, Panama. 84 species, 1897 crowns, 16 dates.}
268
+ }
269
+ ```